17.5.4 Standard Performance Tests
When we need to have a standard performance estimation, one might use SPECint.
SPECint is designed to measure processor speed and compare various hardware.
There is a special company to look after this:
SPEC = Standard Performance Eval. Corp., their website is www.spec.org
What is widely known at the moment is SPECint2006.
SPECint2006 includes 12 tests (all in C/C++)—perlbench, gcc, bzip2, …,
xalancbmk.
One can create performance comparison of embedded devices of processor
overall 12 tests. Then, designed system might be 12-dimensional evaluation of the
known system. Therefore, application segment relative performance might be
known. For more details about standard performance, see
http://www.spec.org/cpu2006/results/cint2006.html.
17.5.5 Real Processor Hardware Comparison
The performance of processor is reduced if instruction uses intensively data traffics
between memory and active zone. Today, a top-of-the-line x86 microprocessor runs
at 3–4 GHz, while the memory runs at 100–500 MHz.
The time to execute an instruction inside a CPU is almost negligible in contrast
with access memory time. In other words, the performance of architecture depends
on memory traffics.
Proposed recently by ITACS performance of ReP is evaluated by the comparisons with 80 Â 88 architecture and MIPs’ instruction architecture.
Simple instructions dominate this list and are responsible for 96% of the
instructions executed. These percentages are the average of the five SPECint92
programs [7].
Various mixes of instruction cover different applications. However, we can
evaluate the CPU performance based on the average case as illustrated in
Table 17.1. Within this evaluation, we assume loading data and storing data into
memory takes equal time.
The execution time comparison between 80 Â 86 and ReP is shown in
Table 17.2.
MIPs dynamic instruction mix for five SPECint2000 programs (gap, gcc, gzip,
mcf, and perl) is presented in [7]. The execution time comparison between MIPs
and ReP is shown in Table 17.3.
Table 17.4 compares the performance of MIPs and ReP.
Modern embedded systems need efficient cost–performance ratio and very often
minimize power consumption. One of the design options is as it was mentioned
above the design as much as possible and therefore reduce the hardware overheads.
236
17 On Performance: From Hardware up to Distributed Systems
When we need to have a standard performance estimation, one might use SPECint.
SPECint is designed to measure processor speed and compare various hardware.
There is a special company to look after this:
SPEC = Standard Performance Eval. Corp., their website is www.spec.org
What is widely known at the moment is SPECint2006.
SPECint2006 includes 12 tests (all in C/C++)—perlbench, gcc, bzip2, …,
xalancbmk.
One can create performance comparison of embedded devices of processor
overall 12 tests. Then, designed system might be 12-dimensional evaluation of the
known system. Therefore, application segment relative performance might be
known. For more details about standard performance, see
http://www.spec.org/cpu2006/results/cint2006.html.
17.5.5 Real Processor Hardware Comparison
The performance of processor is reduced if instruction uses intensively data traffics
between memory and active zone. Today, a top-of-the-line x86 microprocessor runs
at 3–4 GHz, while the memory runs at 100–500 MHz.
The time to execute an instruction inside a CPU is almost negligible in contrast
with access memory time. In other words, the performance of architecture depends
on memory traffics.
Proposed recently by ITACS performance of ReP is evaluated by the comparisons with 80 Â 88 architecture and MIPs’ instruction architecture.
Simple instructions dominate this list and are responsible for 96% of the
instructions executed. These percentages are the average of the five SPECint92
programs [7].
Various mixes of instruction cover different applications. However, we can
evaluate the CPU performance based on the average case as illustrated in
Table 17.1. Within this evaluation, we assume loading data and storing data into
memory takes equal time.
The execution time comparison between 80 Â 86 and ReP is shown in
Table 17.2.
MIPs dynamic instruction mix for five SPECint2000 programs (gap, gcc, gzip,
mcf, and perl) is presented in [7]. The execution time comparison between MIPs
and ReP is shown in Table 17.3.
Table 17.4 compares the performance of MIPs and ReP.
Modern embedded systems need efficient cost–performance ratio and very often
minimize power consumption. One of the design options is as it was mentioned
above the design as much as possible and therefore reduce the hardware overheads.
236
17 On Performance: From Hardware up to Distributed Systems